This startup wants to change how mathematicians do math

Axiom Math, a Palo Alto startup, has released Axplorer, a free AI tool designed to democratize high-level mathematical discovery. The tool is a streamlined, accessible version of PatternBoost, a system co-developed by Axiom scientist François Charton during his time at Meta. While PatternBoost required a massive supercomputer cluster, Axplorer runs efficiently on a single Mac Pro, putting its power into the hands of individual researchers. This shift is part of a broader push, including DARPA's expMath initiative, to integrate AI into mathematical exploration.

Axplorer's capability was demonstrated by solving the Turán four-cycles problem—a significant challenge in graph theory—in just 2.5 hours. This matched the result of the original PatternBoost, which ran for three weeks on thousands of machines. The tool works by generating examples based on user input, allowing mathematicians to iteratively select interesting patterns and explore new, uncharted mathematical territories. This approach contrasts with Large Language Models (LLMs) like GPT-5, which Charton notes are "conservative" and better at derivative work based on existing data.

The release addresses a critical access problem in AI-assisted math. Tools like Google DeepMind's AlphaEvolve and the original PatternBoost have shown promise but are locked behind corporate walls or require immense computational resources. Axplorer breaks down that barrier, offering a practical tool for exploratory, experimental mathematics that can lead to foundational breakthroughs with knock-on effects for computer science, next-gen AI, and cybersecurity.